Add some nodejs metrics

Author: ericallam

packages/cli-v3/src/entryPoints/dev-run-worker.ts

@@ -213,6 +213,7 @@ async function doBootstrap() {
       forceFlushTimeoutMillis: 30_000,
       resource: config.telemetry?.resource,
       hostMetrics: true,
+      nodejsRuntimeMetrics: true,
       // Drop all system metrics from dev metrics export
       droppedMetrics: ["system.*"],
     });

packages/cli-v3/src/entryPoints/managed-run-worker.ts

@@ -192,6 +192,7 @@ async function doBootstrap() {
       metricReaders: config.telemetry?.metricReaders ?? [],
       resource: config.telemetry?.resource,
       hostMetrics: true,
+      nodejsRuntimeMetrics: true,
     });
 
     const otelTracer: Tracer = tracingSDK.getTracer("trigger-dev-worker", VERSION);

packages/core/src/v3/otel/nodejsRuntimeMetrics.ts

@@ -0,0 +1,64 @@
+import { type MeterProvider } from "@opentelemetry/sdk-metrics";
+import { performance, monitorEventLoopDelay } from "node:perf_hooks";
+
+export function startNodejsRuntimeMetrics(meterProvider: MeterProvider) {
+  const meter = meterProvider.getMeter("nodejs-runtime", "1.0.0");
+
+  // Event loop utilization (diff between collection intervals)
+  let lastElu = performance.eventLoopUtilization();
+
+  const eluGauge = meter.createObservableGauge("nodejs.event_loop.utilization", {
+    description: "Event loop utilization over the last collection interval",
+    unit: "1",
+  });
+
+  // Event loop delay histogram (from perf_hooks)
+  const eld = monitorEventLoopDelay({ resolution: 20 });
+  eld.enable();
+
+  const eldP50 = meter.createObservableGauge("nodejs.event_loop.delay.p50", {
+    description: "Median event loop delay",
+    unit: "s",
+  });
+  const eldP99 = meter.createObservableGauge("nodejs.event_loop.delay.p99", {
+    description: "p99 event loop delay",
+    unit: "s",
+  });
+  const eldMax = meter.createObservableGauge("nodejs.event_loop.delay.max", {
+    description: "Max event loop delay",
+    unit: "s",
+  });
+
+  // Heap metrics
+  const heapUsed = meter.createObservableGauge("nodejs.heap.used", {
+    description: "V8 heap used",
+    unit: "By",
+  });
+  const heapTotal = meter.createObservableGauge("nodejs.heap.total", {
+    description: "V8 heap total allocated",
+    unit: "By",
+  });
+
+  // Single batch callback for all metrics
+  meter.addBatchObservableCallback(
+    (obs) => {
+      // ELU
+      const currentElu = performance.eventLoopUtilization();
+      const diff = performance.eventLoopUtilization(currentElu, lastElu);
+      lastElu = currentElu;
+      obs.observe(eluGauge, diff.utilization);
+
+      // Event loop delay (nanoseconds -> seconds)
+      obs.observe(eldP50, eld.percentile(50) / 1e9);
+      obs.observe(eldP99, eld.percentile(99) / 1e9);
+      obs.observe(eldMax, eld.max / 1e9);
+      eld.reset();
+
+      // Heap
+      const mem = process.memoryUsage();
+      obs.observe(heapUsed, mem.heapUsed);
+      obs.observe(heapTotal, mem.heapTotal);
+    },
+    [eluGauge, eldP50, eldP99, eldMax, heapUsed, heapTotal]
+  );
+}

packages/core/src/v3/otel/tracingSDK.ts

@@ -63,6 +63,7 @@ import {
 import { traceContext } from "../trace-context-api.js";
 import { getEnvVar } from "../utils/getEnv.js";
 import { machineId } from "./machineId.js";
+import { startNodejsRuntimeMetrics } from "./nodejsRuntimeMetrics.js";
 
 export type TracingDiagnosticLogLevel =
   | "none"
@@ -84,6 +85,8 @@ export type TracingSDKConfig = {
   diagLogLevel?: TracingDiagnosticLogLevel;
   resource?: Resource;
   hostMetrics?: boolean;
+  /** Enable Node.js runtime metrics (event loop utilization, heap usage, etc.) */
+  nodejsRuntimeMetrics?: boolean;
   /** Metric instrument name patterns to drop (supports wildcards, e.g. "system.cpu.*") */
   droppedMetrics?: string[];
 };
@@ -326,6 +329,10 @@ export class TracingSDK {
       hostMetrics.start();
     }
 
+    if (config.nodejsRuntimeMetrics) {
+      startNodejsRuntimeMetrics(meterProvider);
+    }
+
     this.getLogger = loggerProvider.getLogger.bind(loggerProvider);
     this.getTracer = traceProvider.getTracer.bind(traceProvider);
   }